Stephen
C.
PeckElectric Power Research Institute, Palo Alto, CA, USA Thomas J. Teisberg
Teisberg Associates, Weston, MA, USA
1. Introduction
In this paper, we use the Carbon Emissions Trajectory Assessment (CETA) Model to investigate the value of information about global warming. The CETA model represents world-wide economic growth, energy consumption, energy technology choice, global warming, and global warming costs over a time horizon of more than 200 years. In CETA, energy technologies and the oil, gas, and coal resource bases are inputs t o an energy submodel, which supplies energy inputs t o a production submodel, and the COz by-product t o the warming submodel. In the production submodel, energy, labor, and capital inputs are used to produce output which is then allocated to con- sumption, investment, energy costs, and damage costs of warming. Because energy costs (and energy-technology choices) are considered together with warming damage costs, the time paths of COz emissions and carbon taxes in our model reflect an optimal balancing of the cost of emission reduction and the benefit of reduced global warming.
2. Parameter Sensitivities
We begin our investigation of the value of information by exploring the sensitivity of optimal policies t o variations in key parameters. We do this a t two points in time: 2030 (before a major transition t o coal based synthetic fuels has occurred) and 2100 (well into this transition). We find that optimal emissions tend t o be insensitive t o parameter variation in 2030, but not in 2100. This implies that resolving uncertainty about these parameters before 2030 is not likely t o have high value - if roughly the same policy is optimal in
'This paper does not represent the views of EPRI or of its members.
this time frame regardless of parameter values, little is gained by resolving uncertainty about these values. Second, and conversely, the sensitivity of optimal emissions in 2100 implies that resolving uncertainty well before 2100 is likely t o have relatively high value.
Our sensitivity results also suggest a subset of key parameters on which we focus in our subsequent value of information analysis. These are pa- rameters which have significant effects on optimal emissions, and which (for technical reasons) affect the benefits rather than the cost of C 0 2 emission control. These key parameters are the warming rate per C 0 2 doubling and the parameters specifying, the warming damage function.
3. Valuing Informat ion
To investigate the value of information about uncertain parameters, we use the paradigm suggested by decision analysis. In this paradigm, information is valued as the difference between (1) the expected value obtained if the state of the world is known before a policy must be adopted, thereby allowing a potentially different policy t o be applied in every possible state of the world, and (2) the expected value obtained if a single policy must be adopted (without knowledge of the state of the world) and then applied across all possible states of the world.
In using this approach, a central issue concerns how the emissions con- trol policy under uncertainty in chosen. In the standard decision analysis approach, this policy is set so as to maximize expected net benefits. How- ever, the global warming problem is being considered in a highly political context involving governments of many countries with differing perspectives and interests. In this context, emissions control policies chosen in the ab- sence of good information may be far from the optimal policy. Thus we present results assuming both that the policy under uncertainty is optimal, and that the policy is arbitrarily chosen in the political process.
There are numerous challenges in valuing information in the context of the global warming problem. First, there is a very large number of uncertain- ties involved in global warming. Second, available assessments of parameter uncertainties are typically limited t o possible ranges at most, while infor- mation on distributional shapes and possible correlations among uncertain parameters is not available. Third, perfect information rarely becomes avail- able all a t once - instead, there is a continuing process of updating "best estimates" over time as information is developed. Finally, even without un- certainty, modeling of global warming is computationally demanding, since
warming involves complex natural and human systems over a time scale measured in centuries.
In the face of these difficulties, we adopt certain simplifications in this paper. First, based on our parameter sensitivity results, we limit our consid- eration t o three key parameters affecting the benefits of emission reductions.
Second, for most of our analysis we treat each parameter in turn as the only uncertain parameter, and represent its probability distribution using three points, with probabilities 116, 213, 116 for the Low case, Central case, and High case, respectively. However, we do conduct an experiment t o explore the implications of joint uncertainty about more than one parameter. For this experiment, we simplify our problem even further by assuming that the two parameters are independently distributed and that these parameters can take on only a High or Low value, each with probability 112. Finally, in all the cases we consider, we assume that information perfectly reveals parameter values.
4. Results Assuming Optimal Policy Under Uncertainty
When we consider single parameter uncertainty assuming that policy under uncertainty is chosen t o maximize expected net benefits, our results suggest that the value of information can be up t o hundreds of billions of dollars.
For the key parameters we consider, we find that the value of information is greatest for information regarding the potential warming anticipated from a given increase in C 0 2 concentration; however, the value of information regarding the future damage costs of warming is nearly as great.
In general, these value of information numbers seem t o justify devoting substantial resources t o resolving global warming uncertainties. Also, since global warming research budgets are now directed primarily a t resolving scientific uncertainties like that about the extent of potential warming, our results provide some support for the position that budgets for research on impacts and adaptation are relatively under-funded, and should be given more resources.
Although resolving uncertainty produces a large benefit relative t o not resolving uncertainty, the benefit of resolving uncertainty quickly is surpris- ingly low. Specifically, we find that the benefit of resolving uncertainty now instead of 20 years from now is roughly 2 percent of the overall benefit of resolving uncertainty. This result is due t o the fact that the optimal energy use policy in our model would be about the same over the next couple of
decades, for any resolution of uncertainty about the key model parameters.
However, by the middle of the next century, optimal energy use policies will become more sensitive t o the key model parameters. Consequently, the ben- efits of accelerating uncertainty resolution by 20 years would be much higher later on.
To obtain a rough sense of the implications of joint uncertainty about two or more model parameters, we conduct an experiment in which we treat two parameters as jointly uncertain and independently distributed. Our results from this experiment are generally consistent with those for single parameter uncertainty. However, there are some noteworthy differences.
First, the value of information about either uncertain parameter is higher when the other parameter is treated as uncertain, rather than treated as known and equal t o its Central case value. Second, the value of resolving uncertainty about both parameters simultaneously is well in excess of the sum of the values of resolving information for each of the two parameters treated as the only uncertain parameter. This result suggests that the sum of the values of information for two or more parameters each treated as the only uncertain parameter understates the value of resolving uncertainty about all those parameters a t once.
5 . Results Assuming Arbitrary Policy Under
Uncertainty
In the forgoing analysis, we assumed that emissions control policy under un- certainty is based on an optimal balancing of the expected costs and benefits of emissions reduction. We also present some results assuming that policy under uncertainty is arbitrarily determined by a real world political process involving the governments of many countries with differing perspectives and interests.
While it is difficult t o forecast what kind of emissions reduction policy might emerge from the political process, whatever policy emerges is unlikely t o be the optimal one. We consider two possible suboptimal policies that might emerge: one is a policy of no emissions reduction before uncertainty is resolved, and the other is a policy of limiting emissions t o the 1990 level until uncertainty is resolved. In either case, we assume that when uncertainty is resolved, the policy will revert t o the optimal one for whatever state of the world is revealed.
When policy under uncertainty is arbitrarily chosen, we find that the value of resolving uncertainty now instead of twenty years from now is much
greater than when policy under uncertainty is optimal. Specifically, if the arbitrary policy under uncertainty were to be no emissions reduction, the benefit of resolving uncertainty is an order of magnitude greater; and if the arbitrary policy were t o be an emissions limit at the 1990 level, the benefit is three orders of magnitude greater.
These results contrast sharply with our earlier ones that suggested there was not a great deal of urgency in resolving global warming uncertainties when an optimal policy is used under uncertainty. Evidently, if early res- olution of uncertainty can head-off implementation of inappropriate C 0 2 control policies, early resolution has huge benefits.
6. Conclusions
The global warming problem is a complicated one, and placing a value on resolution of global warming uncertainties is a difficult task. In this paper, we present a first effort at such an analysis. Obviously, there are important caveats that should be attached to our analysis.
First, the CETA model cannot perfectly represent the future for the next 200 years, even if the key parameters of the model are completely known.
Both the climate model and the economic growth model in CETA are very simple representations of extremely complex systems over a very long period of time; these simple representations necessarily omit many real world factors that bear on the warming problem.
Second, our representation of uncertainty and learning is both limited and simplified. We limit the number of parameters that we treat as uncertain at a given time, we limit the number of possible values that each may take, and we assume that these parameter values are either completely unknown or perfectly known. In addition, the possible values that parameters may take are in most cases just our own estimates of 5 and 95 percent probability points for these parameters.
However, we have conducted a self-consistent exercise t o identify impor- tant driving variables and to estimate the value of information for a selected subset of these variables. Caveats notwithstanding, we believe that our re- sults support the following tentative conclusions:
1. If an optimal policy is used under uncertainty, the value of information is large enough to justify current research efforts, and perhaps t o justify increased emphasis on research into the impacts of warming and cost of adaptation to warming.
2. If an optimal policy is used under uncertainty, ample time is available t o plan and execute a well-designed research program t o resolve uncer- tainties.
3. However, if the political process will choose suboptimal policies and this choice could be prevented by early resolution of uncertainty, the urgency of resolving uncertainty is dramatically increased.